Rational Drug Design for Tubulin Ligands

Researchers: Jean-Claude Bradley and Andrew SID Lang

Objective

We will use rational drug design (both human and bot) to find Ugi products that have the potential to bind to tubulin.

Background

Knowing the hydrogen bonding sites for a known ligand to tubulin (see below) it may be possible to design other ligands using rational drug design. Since the Bradley lab is already experienced in synthesizing Ugi products, we shall focus our design on Ugi products.
2D view of the 3 hydrogen bonding sites between paclitaxel and gamma-tubulin (from PDB ligand viewer )
paclibonds2.jpg
3D view of the 3 hydrogen bonding sites between paclitaxel and gamma-tubulin (from PDB ligand viewer )
pacliHbonds.jpg

Using AutoGrow 2.0 (Andrew Lang)

First Run

The scaffold ugi.smi, see image below, was converted to ugifirstrun.pdb using OpenBabel (GUI) with the option to add polar hydrogens. The ugi.pdb file was then used a the base ligand for AutoGrow 2, a program that runs through a series of docking rounds, growing and mutating the ligands at each round.
ugi.gif
core Ugi scaffold - ugi.smi

AutoGrow was executed using the following command from the AutoGrow_2_0_4/bin directory
/usr/java/jre1.6.0_25/bin/java Main -run_mode Execute -parm_file default.prm
where default.prm is the parameter file:
//DIRECTORIES
working root directory: /home/alang/AutoGrow_2_0_4/run_dir
fragments directory: /home/alang/AutoGrow_2_0_4/fragment/large_fragment
scripts directory: /home/alang/AutoGrow_2_0_4/scripts
 
//INPUT FILES
initial ligand: /home/alang/D-Exp019/ugifirstrun.pdb
receptor: /home/alang/D-Exp019/1JFF.pdb
 
//AUTODOCK PARAMETERS
autodock grid center: -0.5 -16.5 15.0
autodock box size: 14 18 14
 
//EVOLUTION PARAMETERS
number of carryovers: 10 //10
number of children: 20 //20
number of mutants: 20 //20
max number atoms: 150 //500
receptor location: -0.5 -16.5 15.0
receptor radius: 13.4
indices of hydrogens that are not linkers: -1
number of generations: 8 //8
After 6 of 8 generations it was noticed that AutoGrow was only growing and mutating the original Ugi core structure about the single explicit hydrogen in the ugifirstrun.pdb file. This is not very useful from a combiugi viewpoint so the final two rounds were aborted. For interest, we present the two best (most negative affinity) ligands from round 6 - neither compound was found in ChemSpider:
  1. ligand1.out_ligand_1.pdb (SMILES: C(=O)(N1C[NH2]c2[nH]c(nc2C(=NO)NN1c1nc(=O)c2c(nc(cn2)C(=O)O)[nH]1)Oc1c2nc(cnc2nc(N)n1)C(=O)O)C(C)(C)N(C(=O)C)C - Affinity: -10.9)
  2. ligand3.out_ligand_1.pdb (SMILES: c1nc(c(n1Nc1nc2c(nc1C(=O)O)[nH]c(nc2=O)N1N(C(=O)C(C)(C)N(C(=O)C)C)C[NH2]c2[nH]c(nc2C(=NO)N1)Oc1c2nc(cnc2nc(N)n1)C(=O)O)N)C#N - Affinity: -10.8)
The following images are the 2D images rendered using depict, with both SMILES causing the following warning:
 WARNING: Atom has unusual valence 4 (normal 5) (dy_rmbord)
 WARNING: ...C)(C)N(C(=O)C)C)C[NH2]c2[nH]c(nc2C(=NO)N... (dy_rmbord)
 WARNING:                     ^^^^^                      (dy_rmbord)
ligand1.gif
ligand1 - depict
ligand3.gif
ligand3 - depict

The following images are the 3D docked configurations created using AutoDock Tools:
ligand1docked.png
ligand1 - docked

ligand3docked.png
ligand3 - docked

Second Run

The scaffold ugi.smi was protonated using molconvertor (part of Marvin 5.5.0.0 in the bin folder) and the following code:
molconvert -3 pdb:+H ugi.smi -o ugi.pdb
The file created had lines that ended in +0, e.g.
HETATM    1  C   UNK     0      -1.456  -5.067   0.269  0.00  0.00           C+0
The file opened just fine in AutoDock Tools and Marvin but caused AutoGrow to fail (exit) when it tried to create the first set of mutants. All the end of line '+0's were removed from ugi.pdb by using search and replace in a text editor. After they were removed, AutoGrow executed successfully using the same code as above, i.e.:
/usr/java/jre1.6.0_25/bin/java Main -run_mode Execute -parm_file default.prm
but with a slightly different default.prm file - I reduced the maximum number of atoms to 100 (from 150) and added instructions not to link to the polar hydrogen atom connected to the nitrogen (ironically the only hydrogen that was connected to in the first run):
//DIRECTORIES
working root directory: /home/alang/AutoGrow_2_0_4/run_dir
fragments directory: /home/alang/AutoGrow_2_0_4/fragment/large_fragment
scripts directory: /home/alang/AutoGrow_2_0_4/scripts
 
//INPUT FILES
initial ligand: /home/alang/D-Exp019/ugi.pdb
receptor: /home/alang/D-Exp019/1JFF.pdb
 
//AUTODOCK PARAMETERS
autodock grid center: -0.5 -16.5 15.0
autodock box size: 14 18 14
 
//EVOLUTION PARAMETERS
number of carryovers: 10 //10
number of children: 20 //20
number of mutants: 20 //20
max number atoms: 100 //500
receptor location: -0.5 -16.5 15.0
receptor radius: 13.4
indices of hydrogens that are not linkers: 25
number of generations: 8 //8
The following are the top ligands after all 8 rounds of AutoGrow
ligand
affinity
SMILES
generation_8_ligand37.pdb
-11.9
C(C(=O)N1CCC(=O)[C@H]([C@@H](O)c2ccco2)N2[C@@H]3C(=C(C[C@]1(C)C(=O)NCOC1=Nc4c(cccc4)N(C1)O)N2n1c(C)c(cn1)C#N)C(=S)NC(=S)N3)c1cc(=O)[nH]n2c1nnc2
generation_8_ligand1.pdb
-11.6
C(C(=O)N1CCC(=O)[C@H]([C@@H](O)c2ccco2)N2[C@@H]3C(=C(C[C@]1(C)C(=O)NCOC1=Nc4c(cccc4)N(C1)O)N2)C(=S)NC(=S)N3)c1cc(=O)[nH]n2c1nnc2
generation_8_ligand2.pdb
-11.5
C(C(=O)N(Cc1c(C)c2c(cc1)cn[nH]2)C(C)(C)C(=O)N[C@H](OC1=Nc2c(cccc2)N(C1)O)n1c(=O)ccc2c1C(=O)[C@@H](C=C2O)O)[C@@H]1C(=O)c2ccc(=O)[nH]c2C(=O)[C@@H]1O
generation_8_ligand28.pdb
-11.5
C(C(=O)N1CCC(=O)[C@H]([C@@H](O)c2ccco2)N2[C@@H]3C4=C([C@H]([C@]1(C)C(=O)NCn1ncc5c(=N)sc(nc15)S)N1c5c(SC)nnc(c5N[C@H]1[S]=C4NC(=S)N3)S)N2)c1cc(=O)[nH]n2c1nnc2
generation_8_ligand34.pdb
-11.5
C(C(=O)N1CCC(=O)[C@H]([C@@H](O)c2ccco2)N2[C@@H]3C(=C(C[C@]1(C)C(=O)N[C@@H](n1ncc4c(=N)sc(nc14)S)c1cc(=O)n([nH]c1=O)c1[nH]nnn1)N2)C(=S)NC(=S)N3)c1cc(=O)[nH]n2c1nnc2
generation_8_ligand29.pdb
-11.4
C(C(=O)N1CCC(=O)[C@H]([C@@H](O)c2ccco2)N2[C@@H]3C(=C(C[C@]1(C)C(=O)NCn1ncc4c(=N)sc(nc14)S)N2)C(=S)NC(=S)N3)c1cc(=O)[nH]n2c1nnc2Sc1c2cn[nH]c2nnn1
generation_8_ligand25.pdb
-11.3
C(C(=O)N1CCC(=O)[C@H]([C@@H](O)c2ccco2)N2[C@@H]3C(=C(C[C@]1(C)C(=O)N[C@@H](n1ncc4c(=N)sc(nc14)S)c1c(C)[nH]c(n1)N(=O)=O)N2)C(=S)NC(=S)N3)c1cc(=O)[nH]n2c1nnc2
generation_8_ligand3.pdb
-11.3
C(C(=O)N1CCC(=O)[C@H]([C@@H](O)c2ccco2)N2[C@@H]3C(=C([C@@H]([C@]1(C)C(=O)NCOC1=Nc4c(cccc4)N(C1)O)Nc1nc(c4ncsc4n1)O)N2)C(=S)NC(=S)N3)c1cc(=O)[nH]n2c1nnc2
generation_8_ligand5.pdb
-11.2
C(C(=O)N(C)[C@](C)(CSc1nc2c(nc(n2n2c3c(c[nH]n3)c(=O)nc2S[C@@H]2c3ncnc(c3N=N2)O)O)c(n1)O)C(=O)NCOC1=Nc2c(cccc2)N(C1)O)[C@@H]1C(=O)c2ccc(=O)[nH]c2C(=O)[C@@H]1O
generation_8_ligand4.pdb
-11.2
C(C(=O)N(CNC(=O)c1c[nH]nc1NN)C(C)(C)C(=O)N[C@H](OC1=Nc2c(cccc2)N(C1)O)n1c(=O)ccc2c1C(=O)[C@@H](C=C2O)O)[C@@H]1C(=O)c2ccc(=O)[nH]c2C(=O)[C@@H]1O
None of the ligands above are in ChemSpider and I assume they would be difficult to synthesize. The ligands are also large. I would suggest using AutoGrow when the expected ligand is not too much bigger than the initial starting core (ligand). Using AutoGrow in this case seems to produce viable ligands but ligands that would probably be as difficult to synthesize as paclitaxel itself. [Yes this is definitely a weakness of using an AutoGrow approach - even if you find a compound with a good theoretical fit it will almost certainly be difficult to synthesize. Since paclitaxel is such a large molecule I think we would have better luck creating a Ugi library with larger commercially available starting materials. JCB]

Using a Mathematician's Brain (Andrew Lang)

Examining the hydrogen bonding sites (see the first two images) of paclitaxel I kept the first two hydrogen bond sites and made a combi-Ugi set of virtual compounds that will be docked against tubulin. These products should be easier to synthesize that the ligands found by AutoGrow, the paclitaxel carboxylic acid is available for purchase CSID: 2043006.
Building the Library
The first two hydrogen bonding sites can be preserved if the following carboxylic acid is used in the Ugi reaction:
c1ccccc1C(=O)NC(c2ccccc2)C(O)C(=O)O

This will form a Ugi product with the following structure:
c1ccccc1C(=O)NC(c2ccccc2)C(O)C(=O)N([R1])C([R2])C(=O)N[R3]
ugiwithRs.png
Where the other components of the Ugi reaction are:
amine: [R1]-NH2
aldehyde: [R2]-C=O
isocyanide: [R3]-N#C

Using the same set of amines, aldehydes, and isocyanides that were used to generate UClib007 and that are readily available in the Bradley lab, we generated UClib008 using the carboxylic acid side chain of paclitaxel. The generated Lib008 Ugi Product SMILES were then converted to docking-ready (fully protonated) .pdb files using molconverter and the following code:
molconvert -3 pdb:H lib008.smi -m -o lib8ligand.pdb
Results
Lib008 was docked against Tubulin on ORU's cluster (it took several days to run). The results were analysed and the top 21 ligands were docked locally using PaDEL-ADV (3 hours run time). Docking scores were then averaged and the top 7 docked ligands (pdb - zip) are presented below.
DExp019AllThreeDocked.png
Top Three Docked Ligands - Note the paclitaxel acid is not at the top for the last two ligands

ID
amine
aldehyde
isocyanide
product
AveScore
1482
NC1=C(Cl)C=CC=C1
O=Cc2cc3c(c1c2cccc1)cccc3
[C-]#[N+]C(S(c1ccc(cc1)C)(=O)=O)
O[C@H]([C@@H](NC(=O)c1ccccc1)c2ccccc2)C(=O)N(C1=C(Cl)C=CC=C1)C(c2cc3c(c1c2cccc1)cccc3)C(=O)NC(S(c1ccc(cc1)C)(=O)=O)
-10.8
1497
NC1=C(Cl)C=CC=C1
O=Cc1ccccc1C
[C-]#[N+]C(S(c1ccc(cc1)C)(=O)=O)
O[C@H]([C@@H](NC(=O)c1ccccc1)c2ccccc2)C(=O)N(C1=C(Cl)C=CC=C1)C(c1ccccc1C)C(=O)NC(S(c1ccc(cc1)C)(=O)=O)
-10.2
542
NCc1ccccc1
O=CC=CC
[C-]#[N+]C(S(c1ccc(cc1)C)(=O)=O)
O[C@H]([C@@H](NC(=O)c1ccccc1)c2ccccc2)C(=O)N(Cc1ccccc1)C(C=CC)C(=O)NC(S(c1ccc(cc1)C)(=O)=O)
-10.15
543
NCc1ccccc1
O=CC=CC
[C-]#[N+]C1CCCCC1
O[C@H]([C@@H](NC(=O)c1ccccc1)c2ccccc2)C(=O)N(Cc1ccccc1)C(C=CC)C(=O)NC1CCCCC1
-10.05
742
NC1CCCCC1
O=Cc1ccc(C)cc1C
[C-]#[N+]C(S(c1ccc(cc1)C)(=O)=O)
O[C@H]([C@@H](NC(=O)c1ccccc1)c2ccccc2)C(=O)N(C1CCCCC1)C(c1ccc(C)cc1C)C(=O)NC(S(c1ccc(cc1)C)(=O)=O)
-10
307
Nc1ccccc1
O=Cc1ccc(C)cc1C
[C-]#[N+]C(S(c1ccc(cc1)C)(=O)=O)
O[C@H]([C@@H](NC(=O)c1ccccc1)c2ccccc2)C(=O)N(c1ccccc1)C(c1ccc(C)cc1C)C(=O)NC(S(c1ccc(cc1)C)(=O)=O)
-9.95
107
NC1=CC(Cl)=CC=C1
O=CC=CC
[C-]#[N+]C(S(c1ccc(cc1)C)(=O)=O)
O[C@H]([C@@H](NC(=O)c1ccccc1)c2ccccc2)C(=O)N(C1=CC(Cl)=CC=C1)C(C=CC)C(=O)NC(S(c1ccc(cc1)C)(=O)=O)
-9.85
A few interesting things to note here. First, it is surprising that the paclitaxel side chain is not always docked in the same orientation as it is in the full paclitaxel molecule. This could mean that the paclitaxel side chain is not as important as assumed in this analysis, though it could also mean that the docking algorithm is missing something important. Secondly, a big concern was that the Ugi products generated from a standard library (like library007) may not be big enough (paclitaxel is comparatively bigger than the Ugi products that have previous been made in the UsefulChem project). By examining the docking visualization it seems that that assumption may not be true and that library007 may contain potential ligands, e.g. 1482 looks big enough, though the acid used here is itself relatively large and may be worth including in the next version of lib007. Thirdly, TOSMIC is showing up again as it did in the second run of D-EXP018. The sulfonyl group seems particularly well suited to sit between the two hydrogen bonding sites at the top.

Log

2011-05-12
  • Downloaded and installed Ubuntu using the Ubuntu Windows Installer (AutoGrow 2.0 only runs on linux).
  • Downloaded and installed on Ubuntu:
  1. Python (v 2.7.1)
  2. MGLTools (v 1.5.4)
  3. AutoDock Vina (v 1.1.1)
  4. AutoGrow (v 2.0.4)
  • By following the tutorial, set AutoGrow running overnight using the default settings.
2011-05-13
  • Checked progress of AutoGrow and it was 1/4 complete (2 out of 8 runs complete) - left it running.
2011-05-14
  • Checked progress of AutoGrow and it was 1/2 complete (4 out of 8 runs complete), seems to take a little longer to dock now, probably because the molecules are growing in size - left it running.
2011-05-17
  • With 6 of 8 rounds completed, docking was aborted when I noticed that AutoGrow was only attaching new functional groups to one end of the ugi.pbd file.
  • A new fully protonated ugi.pdb file was created using Marvin and AutoGrow was restarted and left to run overnight.
2011-05-28
  • All 8 rounds completed in 11 days.
2011-05-29
  • Began mathematician rational drug design
2011-05-31
  • UClib008 generated and uploaded to wiki.
  • Docking-ready .pdb files for UClib008 were generated.
2011-06-10
  • Results from docking of lib008 on cluster were analysed.